Research On Anomaly Detection Method For Internet Of Vehicles Based On Privacy Protection

The Internet of vehicles(Io V)includes vehicular AD hoc networks(VANETs)in which vehicles communicate with the surrounding environment and the In-vehicle network in which various systems inside the vehicle are interconnected.They provide communication and data services for intelligent connected vehicles.These services cover vehicle information processing,vehicle condition monitoring,vehicle remote control,etc.,in order to achieve intelligent and interconnected vehicles.However,the high degree of mobility of vehicles and the huge scale of connected vehicles make the topology of the Io V change frequently;In addition,the Io V involves a wide range of vehicles,terminals,servers and other devices,and the use scenarios and application scenarios are complex and changeable.Therefore,the Io V is vulnerable to illegal malicious attacks.The anomaly detection method of Io V based on machine learning can detect and identify malicious attacks and help to establish a reliable communication service environment.However,anomaly detection services require the collection and sharing of real vehicle communication data,which poses a risk of vehicle privacy leakage.Aiming at malicious attacks in Io V communication services and privacy leakage problems in anomaly detection services,this thesis conducts Io V anomaly detection methods based on privacy protection,and puts forward two solutions according to the scope of vehicle communication.The specific research work of this thesis is as follows:(1)Aiming at the problem of malicious attacks and privacy leakage faced by VANETs,a cooperative anomaly detection scheme based on Differential privacy for VANETs was proposed.Firstly,based on the MEC service architecture,a distributed GAN network collaborative anomaly detection scheme is proposed,with generators and discriminators deployed on central and edge servers respectively.Secondly,to solve the problem of information leakage in the Collaborative learning process caused by the lack of privacy protection mechanism,a privacy protection method DP-T for discriminator networks is designed.The privacy budget is allocated according to the output of neurons to the model,and the sensitivity is smoothed to reduce redundant noise.Finally,the experimental results indicate that the proposed scheme has stable anomaly detection capability,which reduces the computational and memory burden compared to centralized vehicular self-organizing network anomaly detection.(2)Aiming at the problems of malicious attacks and privacy leakage faced by In-vehicle networks,an anomaly detection scheme based on binary random response was proposed.Firstly,a binary random perturbation algorithm was designed to protect the privacy of onboard data.By dividing the fixed point binary into two parts and assigning different parameters,the availability of perturbation data was ensured.Secondly,based on the temporal nature of the CAN,a convolutional network TCN constructed by relying on spatiotemporal sequence relationships is used to achieve anomaly detection of desensitized vehicle data.Among them,expanding causal convolution accelerates training,shortens detection time,and satisfies the vehicle’s tolerance for communication delay.Finally,the experimental results indicate that the proposed scheme has efficient anomaly detection capabilities in vehicle networks,which reduces time overhead compared to other vehicle network anomaly detection schemes.